Sentiment classification using automatically extracted subgraph features

In this work, we propose a novel representation of text based on patterns derived from linguistic annotation graphs. We use a subgraph mining algorithm to automatically derive features as frequent subgraphs from the annotation graph. This process generates a very large number of features, many of which are highly correlated. We propose a genetic programming based approach to feature construction which creates a fixed number of strong classification predictors from these subgraphs. We evaluate the benefit gained from evolved structured features, when used in addition to the bag-of-words features, for a sentiment classification task.